ludar / elastic-related

Elastic 2.x branch

elastic-related

Basic composer.json to use the package

{
	"repositories": [
	{
		"type": "git",
		"url": "https://github.com/ludar/elastic-related.git"
	}	
	],
	"require": {
		"ludar/elastic-related": "dev-master"
	}
}

The package is a simple wrapper around elasticsearch's more_like_this query. It can be used to implement "related things" stuff based on some text fields.

In the code samples below it is assumed

$client = new ludar\ElasticRelated('my_index', 'my_doc_type');

Sample usage

1. Setup the index

$client->esSetupIndex(['tags']); // we only use one "tags" field for this example

More methods to use along with esSetupIndex():

• bool esIndexExists()
• esDropIndex()

2. Index some documents. IDs can be numbers or strings.

foreach ($docs as $doc) {
	$client->esIndex($doc['id'], [
		'tags' => implode(',', $doc['keywords']),
	]);
}

More methods to deal with indexed documents:

• esDelete()
• elastic_get_result esGet()

3. Ask for related documents

// It returns id => score array sorted by score desc
$related = $client->relatedTo($doc['id'], 'tags', 20);

Multifield scenario

Above is the most simple case: elastic index only consists of a single field.

If we need related stuff based on more than just one field we have two choices:

• either stick to the relatedTo() method like $client->relatedTo($doc['id'], ['field1', 'field2'], 20);. This way it works exactly like more_like_this on multiple fields.
• or use more sophisticated boostedRelatedTo() method which is backed by elastic dis_max query.

Here the difference is.

This time our index has three fields

$client->esSetupIndex(['keywords', 'title', 'description']);

We fill it like this

foreach ($docs as $doc) {
	$client->esIndex($doc['id'], [
		'keywords' => implode(',', $doc['keywords']),
		'title' => $doc['title'],
		'description' => $doc['description'],
	]);
}

And ask for related documents like this

$related = $client->boostedRelatedTo($doc['id'], [
	[
		'fields' => 'keywords',
		'boost' => 4, // boost keywords matches the most
	],
	[
		'fields' => 'title',
		'boost' => 2, // boost title matches some more
	],
	[
		'fields' => 'description',
		//use default boost=1 for description matches
	]
], 20, 1);

The last agrument to boostedRelatedTo() (1 in the example above) is very important. It is mapped to the tie_breaker parameter in elastic dis_max query. It additionally boosts docs with multifield matches. Suitable boosts and tie breaker value are to be found by tries.

Bulk indexing

Indexing docs with esIndex() might take too long. To fasten it up we can use either boring esBulk() method or the hipstered ludar\ElasticBulkIndex wrapper class around it.

Use it like this

$bulk = new ludar\ElasticBulkIndex($client, 100); // 2nd argument is the batch chunk size
foreach ($docs as $doc) {
	$bulk->index($doc['id'], [
	  // Same data you'd use in esIndex()
		'tags' => implode(',', $doc['tags']),
		'title' => $doc['title'],
	]);
}
$bulk->done(); // Finalize it